[wip/poc] impl: use flake for profile packages

[gcurtis]

This is a proof-of-concept that changes devbox so that it uses the flake when installing packages to a profile. The idea is that this moves a lot of logic out of Devbox and into Nix, where things are more reproducible. It also makes the flake the single source of truth -- packages added to the flake automatically get added to the profile.

When running this in a project where most packages are already in the Nix store (and therefore most time is spent syncing the profile), this runs about a 2.3x faster:

$ cd ~/src/axiom
$ rm -rf .devbox
$ time devbox install # with this PR
Ensuring packages are installed.
✓ Computed the Devbox environment.
Finished installing packages.
devbox install  0.25s user 0.16s system 48% cpu 0.837 total
$ rm -rf .devbox
$ time devbox install # current main
Ensuring packages are installed.

Installing 12 packages: go@latest, buf@latest, python310@latest, sops@latest, kubernetes-helm@latest, postgresql@latest, nodejs@latest, yarn@latest, awscli2@latest, golangci-lint@latest, flyctl@latest, gotty@latest.

[1/12] go@latest
[1/12] go@latest: Success
[2/12] buf@latest
[2/12] buf@latest: Success
[3/12] python310@latest
[3/12] python310@latest: Success
[4/12] sops@latest
[4/12] sops@latest: Success
[5/12] kubernetes-helm@latest
[5/12] kubernetes-helm@latest: Success
[6/12] postgresql@latest
[6/12] postgresql@latest: Success
[7/12] nodejs@latest
[7/12] nodejs@latest: Success
[8/12] yarn@latest
[8/12] yarn@latest: Success
[9/12] awscli2@latest
[9/12] awscli2@latest: Success
[10/12] golangci-lint@latest
[10/12] golangci-lint@latest: Success
[11/12] flyctl@latest
[11/12] flyctl@latest: Success
[12/12] gotty@latest
[12/12] gotty@latest: Success
✓ Computed the Devbox environment.
Finished installing packages.
devbox install  0.69s user 0.53s system 62% cpu 1.958 total

So far it passes all tests except for two:

1. The php testscript fails because the packages installed by the plugin get removed from the PATH by devbox. I'm assuming this is fixable.
2. The unfree testscript fails because the new syncing logic doesn't output progress (because it's a poc).